AI Platform Engineer

Elixir OTP Phoenix LLM gateway Agent runtime MCP Anthropic Managed Agents API Provider-agnostic LLM client SSE streaming Prompt registry Structured output Jido Distributed systems Python SQL PostgreSQL Azure AWS Docker CI/CD Git Job Description About the Role This is the engineer who owns the substrate everyone else builds on. While AI engineers ship features, you'll own the runtime, abstractions, and guardrails that let agents and LLM features move from prototype to production without every team rebuilding the foundation. Your success shows up in how much faster other engineers can ship because of what you've built. You'll design and maintain the agent runtime that orchestrates Claude agents in production, including session lifecycle, SSE event streaming with reconnect and de-duplication, crash-safe tool dispatch, and the decision between self-hosted execution, where sensitive data stays on company infrastructure, and managed cloud sandboxes. You'll also own the LLM abstraction layer: a provider-agnostic client for text, structured JSON-schema output, and streaming tool use, plus a versioned prompt registry and model gateway that keeps the company multi-model, with Anthropic native and Bedrock in place today. Standing up that gateway is one of the first concrete projects. The platform is built with Elixir/OTP and Phoenix, alongside Python agent workers, so this is a genuinely polyglot environment. Everything sits behind swappable behaviours with mock implementations for testing. You'll define the reusable patterns, from behaviour-plus-mock interfaces for external services to tool definitions as composable actions, and you'll turn proven workflows into versioned, shareable Skills. Responsibilities Design and maintain the agent runtime, including session supervision, SSE relay with exponential-backoff reconnect, PubSub fan-out, and crash-safe tool dispatch Own the LLM abstraction layer and model gateway, covering the provider-agnostic client, prompt registry, structured-output validation, and the migration path toward native Anthropic endpoints Define the reusable platform patterns other engineers build on, including behaviour-plus-mock seams, composable tools and actions, and shareable Skills Operate the deployment and integration plumbing across background-job orchestration, multi-environment rollout, secret management, and connectors to enterprise systems Decide where work runs, self-hosted workers or managed cloud sandboxes, based on data sensitivity Qualifications Strong systems engineering experience in a functional and/or concurrent language. Elixir/OTP is a plus. Deep functional or distributed-systems experience, plus willingness to ramp, matters most. You should also be comfortable working across an Elixir and Python boundary. Hands-on experience building LLM or agent applications, including tool or function calling, structured output, prompt design, streaming, and the failure modes that come with them Fluency with distributed and streaming systems: supervision, backpressure, retries, idempotency, and graceful degradation Cloud infrastructure and deployment experience in Azure or AWS, along with real testing discipline, including DI, mocks at service boundaries, and behavior-asserting tests A platform mindset. You naturally turn a one-off solution into a reusable interface. Preferred Skills Direct Anthropic API experience, including Claude tool use, the Managed Agents API, MCP, or Skills Familiarity with actor or agent frameworks, SQL analytics, or observability tooling Prior experience building an LLM or model gateway, or a multi-provider routing layer Background in finance, accounting, audit, or another regulated, compliance-sensitive domain Tech Stack (representative) Elixir/OTP and Phoenix LiveView, Python, Anthropic Managed Agents API, Jido, SSE relay and PubSub, Oban, Microsoft Graph, Azure Blob, Salesforce integration fabric, Azure Container Apps, managed identity, Key Vault, and SOC2-grade audit logging. Location: Remote, US-based | Type: 6-month contract-to-hire #J-18808-Ljbffr Elios, Inc.